Provides optional parallel running of repetitions

[meatballs]

• Adds run_serial_repetitions and run_parallel_repetitions methods to Tournament
• Adds new argument to run_tournament (-p or --processes) to define number of parallel processes to run. The default is serial processing. Actual number used is limited to the cpu_count.
• Refactors playing of round robin and updating of result set out of Tournment.play and into separate methods
• The result set update from above now lives within ResultSet.finalise (renamed from ResultSet.init_output)
• Passes logger from TournamentManager to Tournament to log number of processes in use
• Updates test suite to match

And possibly other things I've forgotten.

And possibly not some things I've overlooked!

[meatballs]

On my laptop, running all tournaments serially takes approx. 90s. Running with -p 0 takes this down to approx. 75 seconds

[meatballs]

I'm finished with this one. It's ready to roll.

[drvinceknight]

@JasYoung314 can you also throw your eyes at this one.

[meatballs]

According to various online sources (e.g. http://stackoverflow.com/questions/10037319/how-do-i-tell-if-pythons-multiprocessing-module-is-using-all-of-my-cores-for-ca) python's multiprocessing.cpu_count will return double the actual number of cores for Intel CPUs with hyper threading.

On my Macbook Air, 'About my Mac' shows it has an Intel I7 with 2 cores. cpu_count returns 4.

Running run_tournament with -p 0 launches 4 processes and completes in approx. 75 seconds (as quoted previously). Running with -p 2 reduces this further to 70 seconds.

[JasYoung315]

I've just taken a look over it: Looks great!

It's very similar to a way I had previously implemented parallel processing.

Looking through the code, should line 74 in the tournament.py file be

if self.processes < 2 or self.processes > multiprocessing.cpu_count():

I've left a comment in the commit itself.

[JasYoung315]

Should this be

if self.processes < 2 or self.processes > multiprocessing.cpu_count():

[meatballs]

Looking through the code, should line 74 in the tournament.py file be

if self.processes < 2 or self.processes > multiprocessing.cpu_count():

Yep -that's a typo! Presumably I got away with it simply because it was treated as a property accessor.

[JasYoung315]

It's actually because in python, certain types have a notion of ordering. It's arbitrary but consistent, so you could compare the function cpu_count to the integer self.processes and have the following results

>>> multiprocessing.cpu_count()
4
>>> 5 < multiprocessing.cpu_count
True
>>> 5 > multiprocessing.cpu_count
False

because the type integer is always considered to be less than the type function. More information (and a much more informed explanation can be found here)

Pull request seems awesome!

[meatballs]

Thank you @JasYoung314 for both the explanation and the compliment!!

(Is now a good time to own up to the fact that I'm only really doing this because I've not used Python before and fancied an excuse to learn)?

[JasYoung315]

(I think the time to own up was before you started messing with multiprocessing! Now you're one of us I'm afraid)

[drvinceknight]

Just going through it now: this is very nice to see!

[drvinceknight]

I added something to the README also. Thanks for this @meatballs : huge contribution.

[drvinceknight]

Thanks a lot @JasYoung314 for reviewing this.